Henry Farrell

One thing I have been wondering is if people overindex on Orban's control of media and undervalue his provision of handouts to unhappy rural voters who felt left behind. That has not been Trump's priority and it is harder anyways to carry off in the US system.

The Political Economy of AI: A Syllabus As I’ve noted occasionally before, one of the most potentially useful things that academics do is preparing syllabi, and hence organizing information about the world.

thank you for the recommendation. For the theory that I do, my recent review article www.annualreviews.org/content/jour..., and for the theory that I read, the most recent version of the syllabus that I teach on the political economy of AI www.programmablemutter.com/p/the-politi...

Two weeks until Trump claims the Iran war would never have happened if he’d been president.

if we're on classic British comedies with extended jokes about not mentioning the war www.youtube.com/watch?v=Tms0...

The Village and the Sewer Behind the "Blueskyism" debate

www.programmablemutter.com/p/the-villag...

bactra.org/weblog/918.h...

An official declaration that I am doing oral exams with students (one of the downsides of GenAI!) and then getting on a plane to Europe, and so won't be able to respond to any more of the comments/queries/questions that are proliferating (and are now too many for any human to respond to anyways).

Apple + Microsoft + Eli Lilly paid 46% of all corporate tax to Ireland in 2024
See note from the Ireland's Fiscal Advisory Council.
- www.fiscalcouncil.ie/new-council-...
- www.irishtimes.com/business/eco...

After Software Eats the World, What Comes Out the Other End?

Yes that is exactly the point that the thread is trying to wave its hand towards.

Also I should make it clear that I am _all over_ people disagreeing with me! I have plenty of strongly held opinions, a substantial number of which are likely to be wrong, given the odds. So the underlying loose-ish claim is that we should be disagreeing on different topics in more useful ways.

this is the category error all the AI maximalists make

if I had to break down what I think the impact of AI would be by field, I'd have to first talk about which *subtasks* it's helping people with

the amount of variability by field and task is MASSIVE, ranges from 90% takeover to 0%

I think that’s fair, though I think part of people’s frustration is when it feels like the combativeness isn’t intellectually honest. The level of anger at the Altmans of the world is justified; it doesn’t _need_ inaccuracy to prop it up.

Baa! Baa! All we sheep have gone astray, and I am humbly grateful to you for volunteering to shepherd me back to the fold.

Mind you, in all of what I say (as opposed to Louis) I should stress that I am a non-historian commenting from outside, and what I say should be taken with appropriately sized dollops of salt.

Which on the one hand could be valuable - you _can_ use these technologies to capture large scale cultural phenomena that would otherwise escape, but on the other might end up speedrunning the histoire des mentalites intellectual collapse model if not done with appropriate caution/skepticism.

In other words, I think that similar tradeoffs apply (and obviously, the power critique has its own totalizing tendencies). I do think that there are obvious failure modes - e.g. the 'now we can reconstruct how people thought back then from models trained only on contemporary sources' approach.

So Louis would disagree. His project is about digitizing, but poorly organized 19th century sources with variable and often terrible handwriting. You could mount a Foucauldian critique of the power relations involved, but that is less 'this is useless' than 'maybe useful to the wrong people'

AI’s big rift is like a religious schism, says Henry Farrell In this doctrinal dust-up, very smart people are saying very strange things

yep www.economist.com/by-invitatio...

The Uses of AI for Writing Economic History, Louis Hyman & Matt Jones YouTube video by JRCPPF

If it's me you're disagreeing with, I will enthusiastically stipulate to the 'people who think that code is everything overestimate its general efficacy,' while pushing back some on the suggestion that it is useless for history www.youtube.com/watch?v=ZQga...

Yes, and although it is it genuinely useful application, it feels like the jump from "adding up columns of figures on a calculator" to "Microsoft Excel" rather than the first step on the journey to Mars. The big sceptic case has to be "AI is not an exception to the laws of cost and economics"

Very happy to stipulate to both.

Large Language Models As The Tales That Are Sung Gene Wolfe, Albert Lord, machine culture.

This lays out the argument, riffing on some ideas of Cosma's www.programmablemutter.com/p/large-lang... which suggest to me that there are limits to LLM creativity. Equally, our other coauthor in the _Science_ article has a fascinating new paper pushing a different direction arxiv.org/abs/2601.10825

Large AI models are cultural and social technologies Implications draw on the history of transformative information systems from the past

And I think there is a strong case to be made that programming and poetics have far more in common than you might think at first glance, building on work with @alisongopnik.bsky.social and @cshalizi.bsky.social www.science.org/stoken/autho... about LLMs as "cultural and social technologies."

On the Biology of a Large Language Model We investigate the internal mechanisms used by Claude 3.5 Haiku — Anthropic's lightweight production model — in a variety of contexts, using our circuit tracing methodology.

Interestingly, they actually seem to care about poetry as a use case because it helps to model deep structure! transformer-circuits.pub/2025/attribu... More generally, I take the "country of geniuses in a data center" approach as claiming these technologies are engines of discovery.

At least it's an ethos? The limits of optimal control, from the maximalist and minimalist perspectives

Even if Herbert Simon got a lot wrong about AI, I think his understanding of the variety and complexity of tasks is super useful here. Also, much of @beenwrekt.bsky.social recent writing www.argmin.net/p/at-least-i... and his forthcoming book, which is great. press.princeton.edu/books/hardco...

I think that is true for 95% of commentary on "AI," myself almost certainly included. We abstract from what we know! The other side of this is that if you are in one of the labs, you are inclined to abstract out from programming to believe AI can do a variety of tasks that may/may not be comparable.

equally, while I think this is right, I'm open to finding out that I am wrong as facts and technologies change. We are all trying to get a handle on an extraordinarily weird and difficult-to-understand phenomenon.

Large Language Models As The Tales That Are Sung Gene Wolfe, Albert Lord, machine culture.

My shtick - sort-of implied in this post - www.programmablemutter.com/p/large-lang... - is that it is bad news for jobs that (a) involve the application of pre-baked cultural packages, and (b) can be modeled reasonably successfully in terms of optimization, but that many jobs do not.

it is really good.